Intel’s looming competition with ARM in low-power servers came into clearer focus this week.
Most significant was the announcement by executives with Advanced Micro Devices, Intel’s traditional rival in x86-based processors, that the company will begin making 64-bit ARM chips, as well. AMD will marry the low-power chip designs with the interconnect fabric the company acquired earlier this year when it bought microserver maker SeaMicro, creating chips for dense servers running in hyperscale data center environments.
In addition, ARM, at its TechCon developer conference, unveiled the Cortex-A50 processor series of server chips, based on its ARMv8 architecture. The 64-bit systems-on-a-chip (SoCs) are aimed at a wide range of markets, from smartphones to servers, according to ARM officials.
“The ARM ecosystem will continue its rate of unprecedented innovation to enable diverse platforms,” Simon Segars, executive vice president of ARM’s processor and physical IP divisions, said in a statement. “This will deliver an era of transformational computing, from mobile through the infrastructure and servers that support consumers’ connected, mobile lifestyles.”
In addition, Linaro, an independent engineering group that develops open-source software for the ARM architecture, announced the Linaro Enterprise Group, which aims to create Linux software for ARM-based servers. Joining the group are such high-profile tech companies as AMD, Facebook, Hewlett-Packard and Red Hat. Also among the members are AppliedMicro, Calxeda, Canonical, Cavium and Marvell Technologies. Linaro, which also counts ARM, Samsung, ST-Ericsson and HiSilicon as members, made its announcement at ARM’s TechCon event.
Also at TechCon, Applied Micro demonstrated several open-source applications running on X-Gene SoC based on the 64-bit ARMv8 architecture.
“Our activities today highlight the significant progress that is being made in building out the ARM 64-bit ecosystem, a process that is on-track to perfectly intersect with the availability of our X-Gene silicon,” AppliedMicro President and CEO Paramesh Gopi said in a statement. “We see strong pull from data center and enterprise administrators who require lower power, higher densities and lower total cost of ownership.”
ARM and its partners see a growing demand for high-performance, low-power servers in hyperscale data center environments. Web-based companies like Facebook, Google and Microsoft are running massive, dense data centers designed to run huge numbers of smaller workloads, and need highly energy-efficient servers that can handle those workloads. ARM sees an opportunity in these microservers to chip away at some of Intel’s dominance in the server market.
With its ARMv8 architecture, ARM is including features critical to servers, including 64-bit capabilities, more memory and greater support for virtualization. ARM expects ARMv8-based chips to begin appearing in systems by late 2013 and into 2014.
Top-tier system makers are embracing the idea of low-power ARM-based servers. Hewlett-Packard is partnering with Calxeda to develop such ultra-low-power servers as part of its larger Project Moonshot, while Dell is working with Calxeda and Marvell Technologies on ARM-based servers in its Copper and Zinc initiatives. At TechCon, Dell also showed off a prototype powered by AppliedMicro chips.
The importance of such projects as Linaro’s is in developing the software ecosystem for ARM-based servers.
However, Intel also is pushing its low-power server initiatives, focusing on its Atom platform. The giant chip maker is taking a different tack than AMD, pushing the x86 architecture as the best one for microservers, noting the long history of x86 in servers and the architecture’s support of legacy software. HP officials, while getting a lot of attention for their embrace of Calxeda’s ARM chips, in June opted for Intel’s Atom-based Centerton SoCs for its Gemini microservers, the first systems coming out of Project Moonshot.
AMD, ARM Set Stage for Competition With Intel in Servers
AMD officials for a year have been talking about the company’s ambidextrous strategy of offering third-party processor technology to its customers, so the vendor’s announcement that it would start selling ARM-based server chips in 2014 was of little surprise. AMD in June announced it will incorporate ARM-designed TrustZone security technology into some of its chips.
It also comes at a crucial time for AMD, as it looks to reduce its reliance on the struggling PC market. The company has been financially battered and is laying off hundreds of employees in hopes of stabilizing itself. One of the growth areas officials point to is the dense server space. AMD made a significant move in that direction in February, when it bought SeaMicro for $334 million. A key part of the deal was AMD getting SeaMicro’s Freedom Fabric interconnect technology, which enables the cluster of thousands of servers and offers significant energy-efficiency benefits.
Interconnect fabrics are an increasingly important part of data center infrastructures. Intel also is enhancing its own fabric capabilities, buying QLogic’s InfiniBand business in January and supercomputer maker Cray’s high-performance computing (HPC) interconnect technology in April.
Analysts said AMD’s decision to build ARM-based Opteron server chips to complement their x86 processors makes sense. Charles King, principle analyst for Pund-IT Research, said that while the microserver business may not be a huge slice of the overall server, it’s a significant one, given the players involved.
“While interest in energy-efficient systems tends to wax/wane in broader markets, it has been a core concern among owners of massive, hyperscale data centers for years,” King wrote in a Nov. 1 report. “These companies, including Amazon, Facebook, Google, Microsoft and other leading vendors in cloud, Web 2.0 and related services may be a small if exclusive club, but they represent a potentially huge market opportunity.”
He also noted the presence of companies like Dell, Red Hat and Amazon at AMD’s announcement as an indication of the high level of interest in this market sector.
“[AMD’s] strategy isn’t without risk, since it depends on market and vendor developments that are cast in anything but stone,” King wrote. “But it is also necessary. Without this or similarly bold initiatives, AMD would likely be consigned to also-ran status indefinitely. The right ARM-based solutions, properly exercised, could provide AMD the means of building a far brighter future.”
Roger Kay, principle analyst with Endpoint Technologies Associates, said in a blog post on Forbes.com that between now and when the AMD ARM-based chips will be ready, cloud computing will only continue to grow. In addition, Kay said that while Intel’s argument that ARM chips aren’t as powerful as x86 processors is true, it misses the point of what companies like Facebook and Google are doing.
“[O]ne key aspect of the cloud computing and big data analysis that underlies much of what companies like Google, Facebook, Twitter and Amazon do is its sparse nature,” Kay wrote. “As people with mobile phones and Apple iPads pile photos and tweets onto these platforms, vast data stores are formed that need to be picked over by programs. The picking itself can be quite simple: find the data item, read it and pass it somewhere. But the amount of searching can be enormous.”
It’s this type of work that ARM-based microservers will do well, he wrote.
“Bazillions of tiny cores can all go out at once to the far reaches of the storage pool, looking for the same thing,” Kay said. “When one finds it, it can signal the others to call off the search party. Like an ant colony, the power of this architecture is not in the individual ants but in the colony itself, which, collectively, is much smarter than the individual.
“It’s cheaper, faster, and more efficient to send a whole lot of little cores out at once than to have a few big ones attempt the same task. Thus, once they have 64-bit architecture, dense ARM servers are just the ticket for a fair chunk of tomorrow’s cloud computing.”